In general, programs are executed employing file-access operations to read pertinent data from relevant files and to write the results of a computation task into relevant files. A given task on the K computer is performed using more than 80,000 arithmetic units (compute nodes), and many files may be involved in the computation. If the files accessed are widely dispersed, reading them and writing to them will take a great deal of time. It will also take longer to exchange interim computation results among the compute nodes. Since all this acts as an obstruction to fast computation, we are developing new system software to resolve these problems.